high performance computing on graphics processing units: hgpu.org

This paper presents an automated method for preparing digital X-rays for use by a procedural mesh generator. This process will facilitate the generation of a 3D polygon mesh depicting the bones contained within the X-ray image. The process of preparing the image involves identifying and retaining bone elements whilst removing any superfluous aspects contained within the image, such as text and orientation markers. This will allow a virtual podiatric surgical simulator, VirtuOrtho to feature patient specific bone models generated from the processed X-rays. The algorithm additionally employs parallel processing techniques that are capable of utilizing either multi-core CPUs or a GPU to help to reduce the computational time required to process an image. The relative performance of the multi-core CPU implementation will be compared and contrasted to the performance of the GPU version. As part of the image processing algorithm, two GPGPU algorithms for median filter are proposed: a caching method improves upon the fast, small-radius median filter (McGuire, 2008) and a second method which uses a histogram based technique (Huang et al., 1979) to facilitate filters with large-radius masks to be processed using the GPU.